Scribd
Upload
27 views

Topic20 Sampling Theorem

This document discusses the sampling theorem and its implications for reconstructing continuous-time signals from discrete samples. [1] It states that a continuous signal with maximum frequency ωm can be uniquely determined by samples taken at a rate ωs, as long as ωs is greater than 2ωm. [2] This is known as the Nyquist criterion. [3] Violating it results in aliasing, where high-frequency signal components are misinterpreted as lower frequencies when reconstructed.

Uploaded by

Manikanta Krishnamurthy
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
27 views

Topic20 Sampling Theorem

This document discusses the sampling theorem and its implications for reconstructing continuous-time signals from discrete samples. [1] It states that a continuous signal with maximum frequency ωm can be uniquely determined by samples taken at a rate ωs, as long as ωs is greater than 2ωm. [2] This is known as the Nyquist criterion. [3] Violating it results in aliasing, where high-frequency signal components are misinterpreted as lower frequencies when reconstructed.

Uploaded by

Manikanta Krishnamurthy
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

EE 220 : Signals and Systems

Department of Electronics and Electrical Engineerng


Indian Institute of Technology, Guwahati
Monsoon 2023

Topic 20 : Sampling theorem


Instruction and notes by : Manish

1 Reconstruction of CT signals from its samples


In the previous topic, we observed that the samples of a signal do not always uniquely determine
the corresponding CT signal. For example, consider the figure shown below:

Clearly, in the above case we can not uniquely determine the original signal. Hence, we
need to establish certain constraints to correctly reconstruct the original signal. This is done
with the sampling theorem.

2 Sampling Theorem
CTFT
Statement: Let x(t) ←−−→ X(ω) represent a band-limited signal such that X(ω) = 0 for
|ω| > ωm . If ωs , is the sampling frequency such that ωs > 2ωm , then x(t) can be uniquely
determined by its samples x(nTs ), n = 0, ±1, ±2, ...

In another words, A continuous time signal can be represented in its samples and can be
recovered back when sampling frequency ωs is greater than the twice the highest frequency
component of the original signal.

The minimum sampling rate 2ωm is also called Nyquist sampling rate. The constraint
ωs > 2ωm is called Nyquist criteria and the frequency ωS is called Nyquist frequency. In
many applications, it is more convinient to express frequency in Hertz, hence, the equivalent
constraint is expressed as fS > 2fm , where fs = ωs /2π = 1/Ts .

1
2.1 Derivation
There are multiple ways to derive the sampling theorem criteria. Here, we will look at the
simplest procedure where we can graphically establish the constraint.
Let’s take a CT signal x(t). To generate its sampled version, we multiply it with an impulse
train p(t) having sampling interval TS . The sampled signal can be expressed as
∞
X ∞
X
xδ (t) = x(t)p(t) = x(t)δ(t − nTs ) = x(nTs )δ(t − nTs )
n=−∞ n=−∞

The corresponding Fourier transform of sampled signal can be written as


∞ n=∞
X 1 X
F.T.{ x(nTs )δ(t − nTs )} = X(2π(f − kfs ))
n=−∞
Ts n=−∞

Now, let’s look at its graphical representation:

It’s clear from the figure that to avoid any overlapping of periodic repetitions, we must
follow following constraints:
fs − fm > fm
or we can write that
fs > 2fm (1)
If fs − fm < fm , periodic repetitions will overlap, and it results in aliasing.

2.2 Examples
Example 1: A signal x(t) = cos(2π1000t is sampled at fs = 8000 samples/second rate. Can
we recover this signal back? Write the expression for sampled signal.

Solution: Yes, we can recover the signal back since 8000 > 2 × 1000.
The sampling interval Ts = 1/8000. Let’s set t = nTs = n/8000 to generate the sampled
signal. Thus, sampled signal would be:
π
x(nTs ) = cos(2π1000n/8000) = cos( n)
4

2
Do we need fs > 2fm or fs ≥ 2fm ? Let’s check with an example.

Example 2: A signal x(t) = sin(2π500t is sampled at fs = 1000 samples/second rate. Can


we recover this signal back?

Solution: We can observe that sampling interval fs = 2fm in this example. The sampling
interval would be Ts = 1/1000 s.
Let’s set t = nTs = n/1000 to get the sampled signal:

x(nTs ) = sin(2π500n/1000) = sin(πn) = 0.

Hence, we can’t recover x(t) from x(nTs ) at the given sampling rate.

Note that in many cases it may still be possible to recover the signal at the rate fS = 2fm ,
but this condition is not always true.

Homework:

Hint: In all cases find the maximum frequency component (fm ) of the signal. In part 3, see
topic-4’s example on the convolution of two rectangular pulses to notice how lengths add up
in the convolution of continuous signals.

3
2.3 Reconstructing a signal
Example 3:

Example 4: What happens when fs < 2fm ? : Aliasing

Signal can not be reconstructed completely and we lose some information.

Example 5:

An example of image reconstruction from subsampled vs properly sampled signals. The first
image has lost some high frequency component and hence sharp edges are unclear.

4
3 Quantization
Most signals in life are continuous and for computers to process these continuous signals, they
must be converted to digital form via an Analog-to-Digital Converter (ADC). A digital signal
can be different from its continuous counterpart in two primary ways:
• A signal can be sampled at specific time intervals (Ts ).
• A signal can be quantized at specific voltage levels on y-axis. An exmaple is shown below.

Quantization refers to the process of transforming an analog signal, which has a continuous
set of values, to a digital signal, which has a discrete set.

Quantization error
A digitized sample can have a maximum error of one-half the discretization step size (i.e., ± 21
the least significant bit). Why? Because when we convert an analog value to a digital one, we
round to the nearest integer.

5
Number of levels are defined by the number of bits. For 2 bit quantization, number of levels
(or discreteization steps) = 22 = 4. For n-bit quantization, the number of levels = 2n . Clearly,
we prefer a larger n.

P.S.: The Lenna image is a reminder that tech industry is male-dominated and very biased.
Although this pioneering image contained enough variations in its detail, flatness, frequency
content, shades, and texture that popularized its use as a test image, there is no denying that
it also became popular because men found it attractive. This image has also been called a
symbol of how women have been left out and pushed out of the tech industry. I acknowledge
these issues and as a reminder of our faults, I choose to keep it here.

Reference material
1. Textbook: Signals and Systems by Simon Haykin
2. Lecture notes are inspired from the course materials of JHU 520.214; MIT 6.003; Purdue
ECE-301; UVic ECE-260; and Imperial College E2.5
[Please report any typos in the notes by sending an email to the instructor.]

You might also like